Optional ImageDecoder

[hubgit]

Since upgrading from v4.7.76 to v4.8.69 we've discovered a few issues with ImageDecoder in Chrome (one crash on large images, which is already fixed in Chrome Canary; one faulty image rendering when a JPEG image has an unusual color profile, which I'm going to report).

Might it be possible to have an option to disable the use of ImageDecoder, for situations where we know there could be a problem?

[calixteman]

Would it be possible to share the pdf with the faulty image ?

That said, in considering that ImageDecoder is a pretty recent feature, it needs to be stabilized, so it makes sense to have such an option.
@Snuffleupagus wdyt ?

[Snuffleupagus]

Don't we already have an option for that?

pdf.js/src/display/api.js

Lines 180 to 182 in 13a231c

	* @property {boolean} [isChrome] - Determines if we can use bmp ImageDecoder.
	* NOTE: Temporary option until [https://issues.chromium.org/issues/374807001]
	* is fixed.

[hubgit]

Interesting, I suspect https://issues.chromium.org/issues/374807001 is the issue with large [edit: JPEG] images that I found to have been fixed in Chrome Canary.

Maybe JpegStream.canUseImageDecoder needs to read isChrome too?

[Snuffleupagus]

Given that the existing option is only relevant for BMP, I've submitted PR #19031 to add a more general-purpose option (that can stay even after isChrome is removed).

[hubgit]

The separate issue with the color profile can be avoided by setting colorSpaceConversion: 'none' in the ImageDecoder constructor options, but I suspect that might have unwanted side effects.

[calixteman]

The separate issue with the color profile can be avoided by setting colorSpaceConversion: 'none' in the ImageDecoder constructor options, but I suspect that might have unwanted side effects.

Could you share this specific pdf ?

[hubgit]

The original PDF was private, but I've extracted the problematic image from the PDF, copied the color profile to a new image with exiftool and then converted it to PDF.

In the attached zip file are:

• good.jpg and good.pdf - an unmodified example image, converted to PDF, which has no issues.
• bad.icc - the color profile that's causing a problem.
• bad.jpg and bad.pdf - the example image modified with the bad.icc color profile, converted to PDF, which exhibits the issue when opened in https://mozilla.github.io/pdf.js/web/viewer.html using Chrome.

icc.zip

[hubgit]

I've used this code (with the image served from a local folder) to show that the problem exists when the image is only processed by ImageDecoder:

const res = await fetch('http://localhost:3000/bad.jpg')
const data = await res.arrayBuffer()

const decoder = new ImageDecoder({ 
  type: 'image/jpeg', 
  data,
  // colorSpaceConversion: 'none',
})
const { image } = await decoder.decode()
decoder.close()

const canvas = document.createElement("canvas");
canvas.width = image.displayWidth;
canvas.height = image.displayHeight;
const ctx = canvas.getContext("2d");
ctx.drawImage(image, 0, 0);
document.body.append(canvas);

[calixteman]

The bug has been reported here:
https://issues.chromium.org/issues/378869810

[Snuffleupagus]

Given that there's apparently a few different ImageDecoder bugs specifically in Chrome, should we perhaps (for now) by default set the API-option isImageDecoderSupported = false in Chrome (and simply get rid of the isChrome option)?